Empirical Exercise 3: Daughters and Legislative Voting
London School of Economics and Political Science
November 10, 2025
Anecdote or systematic pattern? Can children influence parental ideology?
❗ How do we establish causation?
Without randomisation, we cannot separate causation from correlation
Balance tests across legislator characteristics
Sample: 434 US House Representatives, 105th Congress (1997-1998)
Child gender operates like nature’s randomised experiment
This suggests daughters make legislators MORE conservative!
But this contradicts:
What’s going wrong?
Conservatives have larger families → More daughters → Negative correlation
More children → More daughters → Conservatives have more daughters
This creates spurious negative correlation between daughters and liberal voting
Mean NOW scores by family composition - 2 children
Democrats and Republicans both show increasing NOW scores with more daughters
Mean NOW scores by family composition - 3 children
More daughters → More liberal voting, holding total children constant
Coefficient on daughters across model specifications
Sign flips from negative to positive when controlling for family size
| Variable | Coefficient | Std. Error | p-value |
|---|---|---|---|
| Number of daughters | 2.72 | (1.05) | 0.010 |
| Total children | -2.83 | (0.76) | 0.000 |
| Republican | -44.87 | (2.11) | 0.000 |
| Female | 10.83 | (2.69) | 0.000 |
| Democratic vote share | 84.16 | (10.87) | 0.000 |
Economically significant effect, not just statistically significant
Heterogeneous treatment effects across subgroups
Effect concentrated among male legislators and Democrats
Robustness across NOW, AAUW, and RTL voting scores
All three measures significant at 5% level
| Index | Organization | Focus | Effect | p-value |
|---|---|---|---|---|
| NOW | National Organization for Women | 20 pieces of legislation | 2.72 | 0.010 |
| AAUW | American Association of University Women | 8 focused issues | 2.63 | 0.019 |
| RTL | National Right to Life Committee | Reproductive rights (reversed) | 4.01 | 0.010 |
Consistent positive effects rule out measurement artifact